[ 
https://issues.apache.org/jira/browse/LUCENE-6780?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Michael McCandless updated LUCENE-6780:
---------------------------------------
    Attachment: LUCENE-6780-heap-used-hack.patch

I was concerned about the peak heap used for "worst case" queries here, so I 
hacked up a quick patch (attached) to test this:

{noformat}
GeoPointInBBoxQuery: field=point: Lower Left: 
[-170.79577068315723,-88.3524701239041] Upper Right: 
[115.75692731020496,51.78004322487766]
  --> 940,015 terms = 34,065,696 bytes

GeoPointDistanceQuery: field=point: Center: 
[-95.87683480747508,-83.99672364681616] Distance: 826889.911703281 meters]
  --> 179,562 terms = 12,446,344 bytes
{noformat}

The patch just records over time the largest number of terms created by the 
query, and then I ran the test for many iterations.

I think this is too high, e.g. too many of these queries in flight at once can 
mean an unexpected OOME.

But I think before we address this we should address the correctness issues 
(the failing seeds for {{TestGeoUtils.testGeoRelations}}).

It could be that to fix these, we place soft limits on how large each query is 
allowed to be?  Meaning, a user who's willing to have more error, willing to 
use more heap, can increase the limit if they want, but by default the limit 
protects the more common use case with smaller shapes.


> GeoPointDistanceQuery doesn't work with a large radius?
> -------------------------------------------------------
>
>                 Key: LUCENE-6780
>                 URL: https://issues.apache.org/jira/browse/LUCENE-6780
>             Project: Lucene - Core
>          Issue Type: Bug
>            Reporter: Michael McCandless
>         Attachments: LUCENE-6780-heap-used-hack.patch, LUCENE-6780.patch, 
> LUCENE-6780.patch, LUCENE-6780.patch, LUCENE-6780.patch, LUCENE-6780.patch
>
>
> I'm working on LUCENE-6698 but struggling with test failures ...
> Then I noticed that TestGeoPointQuery's test never tests on large distances, 
> so I modified the test to sometimes do so (like TestBKDTree) and hit test 
> failures.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org
For additional commands, e-mail: dev-h...@lucene.apache.org

Reply via email to